A108 AUTOMATED DETECTION OF ILEOCECAL VALVE, APPENDICEAL ORIFICE, AND POLYP DURING COLONOSCOPY USING A DEEP LEARNING MODEL

Abstract Background Identification and photo-documentation of the ileocecal valve (ICV) appendiceal orifice (AO) confirm completeness colonoscopy examinations. We hypothesized that an artificial intelligence (AI)-empowered solution could help us automatically differentiate anatomical landmarks such as AO ICV from polyps normal colon mucosa. Purpose aimed to develop test a deep convolutional neural network (DCNN) model can identify AO, these mucosa colorectal polyps. Method prospectively collected annotated full-length videos 318 patients undergoing outpatient colonoscopies. created three non-overlapping training, validation, datasets with 25,444 unaltered frames extracted showing four landmarks/image classes (AO, ICV, mucosa, polyps). For each landmark, we average 30 for time its appearance. All were reviewed by team clinicians. Using quality assessment tool, clinicians examined total 86,754 (7982 8374 32,971 polyps, 37,427 mucosa) verified whether or not frame contained one unique landmark. this research, all white-light colonoscopies, narrow-band imaging excluded. A DCNN classification was developed, validated, tested in separate images. The primary outcome proportion whom AI both them accuracy detecting above threshold 40% (representing value which reliable identification be assumed without increasing false-positive alerts). Result(s) trained on 21,503 recorded 272 patients, validated 1,924 (25 patients) 2,017 (21 frames, respectively. applied transfer learning technique fine-tune parameters endoscopic images using cross-entropy loss function back-propagation algorithm. After training 18 out 21 (85.71%), if accuracies 40%. differentiating 86.37% (95% CI 84.06% 88.45%), 86.44% 88.59%), Furthermore, 88.57% 86.60% 90.33%). Conclusion(s) reliably distinguish It implemented into automated report generation, photo-documentation, auditing solutions improve reporting quality. Please acknowledge funding agencies checking applicable boxes below Other indicate your source funding; MEDTEQ Disclosure Interest M. Taghiakbari: None Declared, S. Hamidi Ghalehjegh Employee of: Imagia Canexia Health Inc. , E. Jehanno T. Berthier L. di Jorio A. N. Barkun Grant / Research support from: co-awardee funded research projects Inc., Consultant Medtronic A.I. VALI Inc, Deslandres: Bouchard: Sidani: Y. Bengio: D. von Renteln ERBE, Ventage, Pendopharm, Pentax, Boston Scientific Pendopharm